The Differential Immune Responses to COVID-19 in Peripheral and Lung

Xu et al. Cell Discovery (2020) 6:73 Cell Discovery https://doi.org/10.1038/s41421-020-00225-2 www.nature.com/celldisc

ARTICLE Open Access The differential immune responses to COVID-19 in peripheral and lung revealed by single-cell RNA sequencing Gang Xu1,FurongQi1, Hanjie Li2, Qianting Yang1, Haiyan Wang1,XinWang1, Xiaoju Liu3, Juanjuan Zhao1, Xuejiao Liao1, Yang Liu1,LeiLiu1,ShuyeZhang 4 and Zheng Zhang 1

Abstract Understanding the mechanism that leads to immune dysfunction in severe coronavirus disease 2019 (COVID-19) is crucial for the development of effective treatment. Here, using single-cell RNA sequencing, we characterized the peripheral blood mononuclear cells (PBMCs) from uninfected controls and COVID-19 patients and cells in paired broncho-alveolar lavage fluid (BALF). We found a close association of decreased dendritic cells (DCs) and increased monocytes resembling myeloid-derived suppressor cells (MDSCs), which correlated with lymphopenia and inflammation in the blood of severe COVID-19 patients. Those MDSC-like monocytes were immune-paralyzed. In contrast, monocyte-macrophages in BALFs of COVID-19 patients produced massive amounts of cytokines and chemokines, but secreted little interferons. The frequencies of peripheral T cells and NK cells were significantly decreased in severe COVID-19 patients, especially for innate-like T and various CD8+ T cell subsets, compared to healthy controls. In contrast, the proportions of various activated CD4+ T cell subsets among the T cell compartment, including Th1, Th2, and Th17-like cells were increased and more clonally expanded in severe COVID-19 patients. ’

1234567890():,; 1234567890():,; 1234567890():,; 1234567890():,; Patients peripheral T cells showed no sign of exhaustion or augmented cell death, whereas T cells in BALFs produced higher levels of IFNG, TNF, CCL4, CCL5, etc. Paired TCR tracking indicated abundant recruitment of peripheral T cells to the severe patients’ lung. Together, this study comprehensively depicts how the immune cell landscape is perturbed in severe COVID-19.

Introduction cases have asymptomatic or mild-to-moderate diseases, Coronavirus disease 2019 (COVID-19) caused by severe around 10% of those infected may develop severe pneu- acute respiratory syndrome coronavirus 2 (SARS-CoV-2) monia and other associated organ malfunctions1. Old age, has been spreading rapidly worldwide, causing serious male sex, and underlying comorbidities are risk factors for public health crisis. Although most SARS-CoV-2-infected causing severe COVID-192; however, the pathogenesis mechanisms remains unclear. Immune perturbations were thought to play crucial roles in the COVID-19 Correspondence: Lei Liu ([email protected])or pathogenesis. Indeed, many reports showed that lym- Shuye Zhang ([email protected])or Zheng Zhang ([email protected]) phopenia and increased blood cytokine levels were closely 1Institute for Hepatology, National Clinical Research Center for Infectious associated with the development or recovery of severe ’ ﬁ Disease, Shenzhen Third People s Hospital, The Second Af liated Hospital, COVID-19 and proposed to treat the patients by inhi- School of Medicine, Southern University of Science and Technology, Shenzhen, 3–6 Guangdong 518112, China biting cytokine storms . 2CAS Key Laboratory of Quantitative Engineering Biology, Shenzhen Institute Recent studies have revealed more aspects of different of Synthetic Biology, Shenzhen Institutes of Advanced Technology, Chinese immune players in COVID-19 infections. Although Academy of Sciences, Shenzhen, Guangdong 518055, China Full list of author information is available at the end of the article SARS-CoV-2 infection in host cells elicits robust secretion These authors contributed equally: Gang Xu, Furong Qi, Hanjie Li

© The Author(s) 2020 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a linktotheCreativeCommons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/. Xu et al. Cell Discovery (2020) 6:73 Page 2 of 14

of chemokines and cytokines, it only weakly produces The clustering analysis showed 29 clusters and 10 major interferons (IFNs)7. In our earlier studies, we showed cell types annotated by marker genes, including T cell + increased recruitment of highly inflammatory FCN1 (CD3D), NK cell (KLRF1), B cell (CD79A), monocyte monocyte-derived macrophages in patients’ broncho- (CD14, FCGR3A), myeloid DCs (mDCs) (CD1C), plas- alveolar lavage fluids (BALFs), suggesting their implica- macytoid DCs (pDCs) (IL3RA), and plasma cells (PCs) tion in the cytokine storm8,9. Intriguingly, T and B cell (IGKC), megakaryocyte (MYL9), cycling cells (MKI67), responses and neutralizing antibodies recognizing SARS- and erythrocyte (HBB) (Fig. 1b, c and Supplementary Fig. CoV-2 were detected among most of the infected patients, S1a, b). By visual inspection, the data integration was with higher levels in those with old age and severe dis- efficient and showed no significant batch effect (Supple- – eases10 12. The dysregulation of other immune cell com- mentary Fig. S1c). Erythrocytes were not included in partments, such as monocytes, dendritic cells (DCs), and subsequent analysis and cycling cells were reclustered into innate-like T cells is also likely present in severe COVID- cycling T, cycling PC, and cycling NK cells based on – 1913 16. specific markers (Supplementary Fig. S1d, e). This dataset To unravel the barely known immune mechanisms indicated significantly dysregulated peripheral immune underlying COVID-19 pathogenesis in an unbiased and landscapes in COVID-19 patients compared to HC, comprehensive manner, we and others have applied the especially among severe cases (Fig. 1d). The most pro- single-cell RNA sequencing (scRNA-seq) to profile the minent changes included an expansion of monocyte and immune cell heterogeneity and dynamics in BALFs, blood, cycling T cells and a reduction of NK, T, and mDC and respiratory tract samples from the COVID-19 populations, thus resulting in largely increased monocyte/ patients. Collectively, those studies revealed a stunted T cell ratios in COVID-19 patients (Fig. 1e, f). The fre- IFN response, depletion of natural killer (NK) and T quency of pDCs was also decreased in severe COVID-19, lymphocytes, loss of major histocompatibility class although the difference was not statistically significant. (MHC) class II molecules, and significantly elevated levels Together, these data show that SARS-CoV-2 infection of chemokine production from monocytes in the greatly perturbs the blood immune cell compartments, – patients8,9,13,14,17 19. However, a complete picture of the particularly in those with severe diseases. COVID-19-induced immune perturbation has not been generated. Here we conducted scRNA-seq analysis of Remodeling of circulating myeloid cell populations in paired blood and BALF samples from the same COVID- patients with COVID-19 19 patients. Our data revealed profound alterations of We observed that the proportions of monocytes were various immune compartments and depicted a dichotomy increased in COVID-19 patients, especially in those with of peripheral immune paralysis and broncho-alveolar severe diseases. To further understand the remodeling of immune hyperactivation in COVID-19 patients. myeloid cell compartment, we re-clustered myeloid cells + and identified five distinct cell types including CD14 + + Results classic monocyte, CD14 CD16 intermediate monocytes, + The overview of dysregulated peripheral immune CD16 non-classic monocytes, DC1, and DC2 (Fig. 2a landscape in severe COVID-19 patients and Supplementary Fig. S2a). The composition of myeloid A high-quality scRNA-seq dataset composed of 200,059 cells in severe COVID-19 patients differed significantly cells were generated that characterized peripheral from that of mild cases and controls. Proportion of + immune cells from three healthy controls (HC), five mild, CD14 monocyte increased significantly in severe and eight severe COVID-19 patients (Fig. 1a). The COVID-19 compared to that in the mild COVID-19 and + metadata of these patients is listed in Supplementary control group, whereas those of CD16 non-classical + + Table S1. Patients with mild diseases were all cured and monocyte (vs controls), CD14 CD16 monocytes (vs discharged after 11–18 days of hospitalization, 2 of the 8 mild COVID-19), and DC2 significantly (vs mild COVID- patients with severe diseases succumbed, whereas other 19 and controls) decreased in severe COVID-19 severe cases recovered after 12–58 days of hospitalization. (Fig. 2b, c). + The clinical course of this patient cohort is similar to CD14 monocytes represent the major peripheral earlier reports, where elderly patients with underlying myeloid cell type, and differential Uniform Manifold diseases were prone to develop severe symptoms and Approximation and Projection (UMAP) projection pat- + showed higher mortality1,2,20. In addition, the plasma terns of CD14 monocytes between COVID-19 and levels of interleukin (IL)-6 and C-reactive proteins (CRP) controls (Fig. 2b) indicated perturbed transcriptome fea- in severe patients were higher, while lymphocyte counts tures. Among the differentially expressed genes (DEGs) in + were reduced, indications of both cytokine storm and CD14 monocytes, we found 116 and 134 upregulated lymphopenia. genes in mild COVID-19 and severe COVID-19 cases Xu et al. Cell Discovery (2020) 6:73 Page 3 of 14

Fig. 1 Single-cell analysis of PBMCs from patients with COVID-19. a The cartoon outlines the study design. PBMC and BALF cells from COVID-19 patients and healthy controls were collected for scRNA-seq characterization using the 10× Genomics platform. The number of samples, analyzed cells, and samples with PBMC and BALF cells simultaneously collected from the same patient are indicated. HC, healthy controls; Mild, mild patients; Severe, severe patients. b The UMAP projection of the combined PBMC scRNA-seq dataset identifies nine major cell types. PC, plasma cells. c The specific markers for identifying each immune cell types in b are indicated. (Pct.Exp. indicates percentage of cells expressed). d Density plots show the UMAP projection of PBMCs from COVID-19 patients and controls. e The bar plot shows the proportions of each cell types in PBMCs from individual subjects. The cell numbers and ratios of monocyte/T cells are listed to the right side. f Comparisons of percentages of each cell types in PBMCs (cycling cells were re-clustered into T and PC subsets) between the two COVID-19 groups and controls (two-sided Student’s t-test, *P < 0.05, **P < 0.01, ***P < 0.001. H, healthy controls; M, mild patients; S, severe patients. compared to controls, vs only 74 upregulated genes in (Supplementary Fig. S2b and Table S2). Gene Ontology comparison between the two COVID-19 groups. In con- (GO) terms of upregulated DEGs included response to trast, we found 217 and 160 downregulated genes in virus, type I IFN, and IFN-γ in both the COVID-19 severe COVID-19 compared to controls or mild cases, groups, and neutrophil activation and energy metabolism respectively, vs only 104 downregulated genes in com- pathways in severe COVID-19 (Fig. 2d). Unexpectedly, parison between mild COVID-19 and controls GO analysis of downregulated DEGs indicated deficient Xu et al. Cell Discovery (2020) 6:73 Page 4 of 14

Fig. 2 Single-cell analysis of peripheral myeloid cell compartments in patients with COVID-19. a UMAP plot of the ﬁve types of myeloid cells in PBMCs. b Density plots show the UMAP projection of peripheral myeloid cells from COVID-19 patients and controls. c Comparisons of percentages of each peripheral myeloid cell types between the two COVID-19 groups and controls (two-sided Student’s t-test, *P < 0.05, **P < 0.01). d Enrichment of GO biological process (BP) terms for upregulated genes (left) and downregulated genes (right) in blood CD14+ monocyte comparisons between mild and HC (M vs H), severe and HC (S vs H), and severe and mild groups (S vs M) (representative terms are shown, adjusted P < 0.01 as indicated by the colored bar). e The heatmaps show the selected differentially expressed genes and their associated GO terms as indicated in d (logFC > 0.41 or < –0.41, adjusted P < 0.01). Mean.Exp, average expression. f Density plots show the composite MHC II signature scores and calprotectin signature scores of peripheral CD14+ monocytes in 2D maps. The horizontal and vertical lines separating the four quadrants represent the median scores of all CD14+ monocytes. The percentages of cells in each quadrant are indicated. g Left panel shows the representative ﬂow cytometric data of HLA-DR expression on CD14+ and CD14– PBMCs. Right plot shows the summarized data from more subjects (two-sided Student’s t-test). h The Pearson correlation of “MDSC-like signature score” and plasma CRP, IL-6 levels, blood neutrophil, CD3+, CD4+, and CD8+ T cell counts. Xu et al. Cell Discovery (2020) 6:73 Page 5 of 14

monocyte functions mainly in severe COVID-19 cases, Abnormally activated lung monocyte-macrophages in such as decreased type I IFN production, cytokine severe COVID-19 secretion, chemokine production, and antigen processing Recently, we discovered the aberrant activation of BALF and presentation (Fig. 2d). We further examined the monocyte-macrophages in severe COVID-198,9. Here, to DEGs associated with those GO terms. Indeed, many further understand the connection between the lung canonical IFN-stimulating genes (ISGs), including ISG15, monocyte-macrophages and their blood counterparts and IFITM1, IFITM3, MX1, IRF7, IFI27, etc., were expressed assess their differential roles in COVID-19, we studied at higher levels in COVID-19 patients than controls, while paired BALF and blood samples from two mild and five the genes associated with neutrophil activation, including severe patients. The integration analysis of BALF and S100A8, S100A9, S100A12, CLU, RNASE2, etc., were circulating myeloid cells showed clusters of neutrophil expressed at higher levels in severe COVID-19 than mild (FCGR3B), mDC (CD1C), monocyte-macrophages (CD14, COVID-19 and controls (Fig. 2e). Despite of upregulated FCGR3A, and CD68) (Fig. 3a and Supplementary Fig. ISGs, we failed to detect higher production of type I or S3a). Macrophage subset classification markers, including type III IFNs in COVID-19 than controls. Regarding the FCN1, SPP1, and FABP4, were differentially expressed by downregulated geneset, MHC II molecules, including circulating and BALF monocyte-macrophages from HLA-DQA1, HLD-DRA, HLA-DRB1, HLA-DMB, HLA- patients with mild or severe COVID-19 (Fig. 3b). Analysis DMA, etc., and cytokine/chemokine genes, including of differentiation trajectory of circulating and BALF IL1B, TNF, CCL3, CCL4, and CXCL8 were expressed at monocyte-macrophages from the same patient revealed a lower levels in COVID-19 patients, especially those with consensus blood-toward-BALF course (Fig. 3c and Sup- severe diseases (Fig. 2e). Thus those upregulated DEGs in plementary Fig. S3b), consistent with the recruitment of + CD14 monocytes from COVID-19 patients reflect the peripheral monocytes into inflammatory tissues as immune response to SARS-CoV-2 infection, while the expected, although the effect of lung immune micro- + downregulated DEGs in CD14 monocytes from patients environment on the exact differentiation trajectory still with severe COVID-19 suggest an immune paralyzed requires investigation and validation. status of those cells. Next, we performed transcriptome analysis of circulat- Myeloid-derived suppressor cells (MDSCs) are a ing and BALF monocyte-macrophages to understand population of heterogeneous immature myeloid cells their functional status. Among the DEGs, there were expanded during inflammatory conditions and could 524 shared upregulated genes and 501 downregulated suppress T cell responses21,22. In peripheral blood, genes in BALF monocyte-macrophages vs those in blood, + monocytic MDSCs have the phenotype CD14 HLA- identified from both mild and severe COVID-19 patients – DR /lo, whereas monocytes are HLA-DR positive23,24. (Fig. 3d and Supplementary Table S3). Such a large Downregulation of MHC II molecules, increased calpro- number of DEGs suggested that significant difference tectin (S100A8 and S100A9), and immune-suppressive existed between the peripheral and lung monocyte- functions are reported features of MDSCs. Indeed, by macrophages. Indeed, GO analysis revealed broad acti- their unique composite scores of MHC II molecules vation of multiple immune pathways in BALF monocyte- (lower levels) and calprotectin (higher levels) vs those macrophages, including response to IFNs and cytokines, scores in mild COVID-19 and controls, we identified that neutrophil activation, and leukocyte migration, while the the monocytes in severe COVID-19 highly resembled pathway involved in myeloid cell differentiation, ATP + MDSCs (Fig. 2f). Decreased levels of HLA-DR in CD14 metabolism, etc. were enriched in blood monocytes (Fig. monocyte from severe patients were further validated by 3e). In addition, these comparisons revealed perturbed flow cytometry (Fig. 2g) and also reported by other stu- pathways in BALF monocyte-macrophages relevant to dies14,18. Intriguingly, MDSC-like scores in our study severe COVID-19, including responses to hypoxia, high positively correlate with serum CRP, IL-6 levels, and temperature, metal ion, wounding, and Fc receptor sig- neutrophil-to-lymphocyte ratio and negatively correlate naling pathways were especially upregulated (Fig. 3e), + + + with decreased blood CD3 , CD4 , and CD8 T cell while pathways related to alveoli macrophage functions counts (Fig. 2h). were downregulated, including lipid metabolism, apop- Together, our scRNA-seq characterization revealed a totic cell clearance, and antigen presentation (Fig. 3e). multifaceted remodeling of peripheral myeloid compart- The representative DEGs involved in those pathways are ment in COVID-19 patients. While the circulating mono- shown in Supplementary Fig. S3c. cytes in COVID-19 patients are featured by heightened ISG Monocyte-macrophages were thought to play key roles responses, they produce little IFNs, cytokines, and chemo- in driving the cytokine storm underlying the development kines. Furthermore, the loss of DCs and emergence of of severe COVID-1925. Therefore, we examined the MDSC-like monocyte suggest their involvements in the cytokine and chemokine levels in monocyte-macrophages immune paralysis of severe COVID-19 patients. in paired blood and BALF samples from the same patient. Xu et al. Cell Discovery (2020) 6:73 Page 6 of 14

Fig. 3 Abnormally activated lung monocyte-macrophages in severe COVID-19. a The myeloid cell data from two mild and ﬁve severe COVID-19 patients who had paired PBMC and BALF samples were integrated and presented on the UMAP. b The expression of monocyte-macrophage markers FCN1, SPP1, and FABP4 were projected to UMAP from a. PM, peripheral cells of mild cases; PS, peripheral cells of severe cases; BM, BALF of mild cases; BS, BALF of severe cases. c Differentiation trajectory of the blood monocytes and BALF monocyte-macrophages from two representative COVID-19 patients, analyzed independently. d Venn diagram shows the number of upregulated and downregulated DEGs in monocyte-macrophage comparisons as indicated. logFC > 0.41 or < –0.41, adjusted P < 0.01. e Enrichment of GO biological process (BP) terms for upregulated genes (left) and downregulated genes (right) in monocyte-macrophage comparisons as indicated. Selected terms are shown, adjusted P value is indicated by the colored bar. f Heatmaps show the expression of selected interferon, cytokine, and chemokine genes in paired blood and BALF monocyte- macrophages derived from the same patient. Stars indicate that the genes are differentially expressed in BALF monocyte-macrophages between mild and severe COVID-19. Purple and green stars show that the gene expression are signiﬁcantly upregulated in severe COVID-19 and mild COVID-19 groups, respectively (MAST; P < 0.01). B, BALF samples; P, PBMC samples. g The levels of selected cytokines and chemokines in paired BALF and plasma samples were measured by CBA (two-sided Wilcoxon test between BALF and PBMC of severe patients).

We found that all types of IFNs (IFNAs, IFNB, IFNG,and data suggest that, although monocyte-macrophages in IFNLs) were minimally expressed by monocyte-macro- BALF are activated, they are immune silent in periphery, phages, whereas cytokines (IL1A, IL1B, IL1R2, IL1RN, IL18, possibly due to continuous engagement of viral stimulation IL6, TNF, IL10,andTGFB1) and multiple chemokines were or the pro-inﬂammatory milieu in the lung. highly expressed in monocyte-macrophages from BALFs Intriguingly, we observed relatively higher levels of anti- but not in those from paired blood samples (Fig. 3f). These inﬂammatory cytokines (IL1R2, IL1RN, and TGFB1) and Xu et al. Cell Discovery (2020) 6:73 Page 7 of 14

lower levels of IL18 in BALF monocyte-macrophages cycling T cells, were significantly increased in severe from severe COVID-19 than mild cases, whereas classical COVID-19 than those in mild COVID-19. The percen- pro-inflammatory cytokines (IL1A, IL1B, IL6, and TNF) tages of CD4-GATA3 and CD4-CCR6 also showed an were comparable between the two groups (Fig. 3f). In increasing trend in severe COVID-19 (Fig. 4d). In addi- contrast, as shown in our earlier studies, monocyte- and tion, single-cell TCR (sc-TCR) analysis revealed increased + + neutrophil-recruiting chemokines (CCL2, CCL3, CCL4, clonal expansion levels in several CD4 but not CD8 T CCL7, CCL8, CXCL1, CXCL2, CXCL3, and CXCL8) were cell subsets in severe COVID-19 than those in mild cases highly expressed, whereas chemokines (CXCL9 and (Fig. 4e, f). Consistently, TCR sharing analysis among CXCL16) recruiting T cells were less expressed by different T cell subsets revealed much more active inter- + monocyte-macrophages in BALFs of severe COVID-19 change between different CD4 T cell subsets and cycling + than those in mild cases (Fig. 3f). The higher levels of T cells but not among CD8 T cell subsets during severe cytokines (IL-1β, IL-6, etc.) and IL-8 in BALFs than paired COVID-19 (Fig. 4g). Together, our data revealed the + plasma was further validated at the protein levels, parti- preferential activation of CD4 T cell responses but sig- + cularly exemplified by the extremely high levels of IL-8 in nificant depletion of multiple innate-like T cell and CD8 BALFs (Fig. 3g). Thus these paired analyses revealed the T cell subsets in peripheral blood as the featured T cell involvement of tissue monocyte-macrophages in cytokine perturbations in severe COVID-19. To further explore the + storms during severe COVID-19, particularly through clues of CD8 T cell lymphopenia, we conducted tran- producing chemokines and recruiting more monocytes scriptome comparisons of MAIT, CD8-GZMK, and CD8- and neutrophils, but less likely attributed to excessive GZMB subsets between the patients and control groups production of pro-inflammatory cytokines. (Supplementary Fig. S4c and Table S4). There was no evidence of T cell exhaustion, activation of cell death The dysregulated peripheral T cell compartments in pathways, and cytokine productions in those cells from COVID-19 patients COVID-19 patients, although the pathways related to NK and T lymphocytes are important antiviral immune virus infection and IFN responses were identified (Sup- cells, which are depleted in severe COVID-1913,26.To plementary Fig. S4d). Similar findings were also noticed further understand the dysregulated NK and T cell by other groups,16 thus exhaustion and cell death are compartments, we re-clustered those cells and identified unlikely the major causes for T cell loss during 18 subsets (Fig. 4a and Supplementary Fig. S4a). NK cells COVID-19. highly expressed KLRF1, KLRC1, and KLRD1. Cycling T cells expressed MKI67. Innate-like T cells included Tracking T cell function and migration across peripheral MAIT (SLC4A10), γδ T(TRGV9), and NKT cells (CD3E, blood and BALFs + KLRF1). CD4 T cells included CD4-Naive (CCR7, SELL), We sought to study T cell movement from blood to CD4-LTB, CD4-GZMK, CD4-GATA3 (Th2), CD4-CCR6 BALFs in paired samples from COVID-19 patients. First, (Th17), CD4-ICOS (Tfh), CD4-GZMB, Treg-SELL, and we integrated data of NK and T cells from peripheral + Treg-CTLA4 subsets, whereas CD8 T cells included blood mononuclear cells (PBMCs) and BALF. Cells were CD8-Naive (CCR7, SELL), CD8-LTB, CD8-GZMK, and re-clustered into nine major types, including NK cells, CD8-GZMB subsets. Pseudo-time trajectory analysis was MAIT, CD4-Naive, CD4-Tm, Treg, CD8-Naive, CD8-Tm, + performed to infer lineage relationship among the CD4 CD8-IL7R, and cycling T cells (Fig. 5a and Supplementary + + and CD8 T cell subsets. Paired T cell receptor (TCR) Fig. S5a). PBMCs included plenty of naive CD4 and + clonotype analysis revealed increased clonal expansion CD8 T cells, while there were fewer naive T cells in along the inferred trajectories (Fig. 4b). BALFs, which were mainly composed of NK cells, CD4- Cell density UMAP projections revealed an obviously Tm, CD8-Tm, and cycling T cells (Fig. 5b). We performed perturbed T cell landscape in severe COVID-19 compared gene expression analysis to determine the functional to the mild COVID and control groups (Fig. 4c). Within divergence of NK and T cells across the peripheral blood the T cell compartment, percentages of innate-like T cells, and BALFs (Supplementary Table S5). Compared with including MAIT and NKT cells, were significantly lower peripheral counterparts, we observed that response to in severe COVID-19 than those in mild COVID-19. Per- virus, type I IFN, and IFN-γ were commonly activated in centages of CD8-Naive, CD8-GZMK, and CD8-GZMB NK, CD4-Tm, and CD8-Tm cells from BALFs (Supple- subsets were also lower in severe COVID-19 patients than mentary Fig. S5b). We also noticed higher levels of those in mild cases, although the difference in CD8- cytokines including IFNG, TNF, CSF1, TNFSF10, and GZMB comparison was not statistically significant (Fig. TNFSF13B; chemokines including CCL3, CCL4, and 4d and Supplementary Fig. S4b). In contrast, the per- CCL5; and IL-15 signaling modules including IL15RA, + centages of several CD4 T cell subsets, including CD4- IL2RB, IL2RG, JAK1, and STAT3 in BALF cells (Fig. 5c). Naive, CD4-LTB, CD4-ICOS, Treg-CTLA4, as well as However, the levels of other T cell relevant cytokines, Xu et al. Cell Discovery (2020) 6:73 Page 8 of 14

Fig. 4 Single-cell analysis of peripheral NK and T cell compartments in patients with COVID-19. a UMAP plot of the 18 subsets of NK and T cells in PBMCs. b Pseudo-time differentiation trajectory of the peripheral CD4+ and CD8+ T cell subsets performed by slingshot. The bar plots in the corner shows the percentages of clonally expanded cells in each T cell subsets. c Density plots show the UMAP projection of peripheral NK and T cells from COVID-19 patients and controls. d Comparisons of percentages of each peripheral NK and T cell types between the two COVID-19 groups and controls (two-sided Student’s t-test, *P < 0.05, **P < 0.01, ***P < 0.001). e Clonally expanded T cells from COVID-19 patients and controls are projected into UMAP from a. f Clonal expansion indexes of T cell subsets from COVID-19 patients and controls are separately displayed. g T cell state transition status among any two clusters is inferred by their shared TCR clonotypes. Each T cell cluster is represented by a unique color. The numbers above the bar indicate the percentages of cells sharing TCRs in those two clusters. including IL4, IL5, IL10, IL13, IL17A, IL17F, IL21, IL25, proportion of BALF T cells belonging to CD8-Tm, CD8 and IL33, were not detected in either blood or BALFs CTL, and cycling T cells subsets could trace their clo- (Supplementary Fig. S5c). notypes back to the paired blood counterparts, especially Next, we utilized the TCR clonotype information to in those patients with severe COVID-19 (Fig. 5f). Here track the migrating T cells in blood and BALF compart- only two mild cases (M1 and M2) had paired blood and ment. The T cell clonal expansion status and clonotype BALF samples for the analysis, and it showed a lower sharing were assessed in patients with paired blood and degree of clonotype shared across the two compartments BALF samples (Fig. 5d, e). Interestingly, considerable (Fig. 5f). The higher levels of chemokines from the lung of Xu et al. Cell Discovery (2020) 6:73 Page 9 of 14

Fig. 5 Tracking T cells across peripheral blood and BALFs in patients with COVID-19. a The T cell data from two mild and five severe COVID-19 patients who had paired PBMC and BALF samples were integrated and presented on the UMAP. b The percentages of each T cell subset in paired BALF (B) and PBMC (P) of the same patient are compared. c Heatmaps display the selected DEGs in NK, CD4-Tm, and CD8-Tm cells in BALF (B) or PBMC (P) from the mild (M) or severe (S) COVID-19 patients. Cytokine-related genes are red marked (logFC > 0.41, adjusted P < 0.01). d The migration index in each T cell subset across paired PBMC and BALF from seven patients are shown (STARTRAC-migr indices). e TCR clonotypes were classified into five different types as indicated by different color bars (singleton indicates non-expanded TCR clonotype, multiplet indicates expanded TCR clonotype, dual-clone indicates those clonotype shared in paired PBMC and BALF samples). The bar plots show the percentages of different types of TCR clonotypes in different T cell subsets from paired PBMC and BALF samples. f The circus plot shows the degree of TCR clonotype sharing across different T cell subsets in PBMC and BALF from the mild and severe COVID-19 groups. g Heatmap shows the selected DEGs in each T cell clones derived from the top 13 TCR clonotypes shared across PBMC vs BALF compartments (logFC > 0.41, adjusted P < 0.01). h The V, J genes of the TCR α and β chains of the top 13 dual clonotypes are listed, and the amino acid sequences of their CDR3 are shown. Xu et al. Cell Discovery (2020) 6:73 Page 10 of 14

severe COVID-19 is likely to lead to increased T cell S100A8, S100A9, and S100A12, were expressed at higher infiltration. To assess the functional adaptation of levels in monocytes from severe COVID-19 patients than migrating T cells, we performed transcriptional analysis those in mild cases. Interestingly, these upregulated and between blood vs BALF T cells derived from the same T downregulated genes are makers of monocytic MDSCs cell clonotypes (Fig. 5h). We found increased expression whose frequencies are known to increase during various of ISGs, CCL4, CXCR4, CXCR6, CD69, GNLY, etc. in inflammatory conditions24,25. Consistent with their BALF cells than those in blood, consistent with an acti- potential immune-suppressive status, we also noticed that vated and tissue-resident phenotype (Fig. 5g and Supple- genes related to cytokine and IFN production were mentary Table S6). Together, these data revealed the downregulated as well. Thus, contradicting with the active recruitment and peculiar activation of peripheral inflammatory role by peripheral myeloid cells in severe T cells in human lungs infected by SARS-CoV-2. COVID-19, the loss of mDCs, emergence of MDSC-like monocytes, and reduced cytokine production actually Discussion suggested a peripheral immune paralysis. Since the The scRNA-seq has been recently been applied to study MDSC-like scores associated with lymphopenia and host immune response in COVID-19 by us and oth- inflammation markers, we speculate a crucial role of – ers8,9,13,17 19. Although these studies helped to reveal MDSC-like monocytes in dampening immune response several aspects of the COVID-19 pathogenesis, a complete and amplifying COVID-19 pathogenesis. Indeed, similar picture has not yet been generated. Here we integrated results and conclusions were recently published by several – scRNA-seq analysis of paired blood and BALFs and independent studies28 31 when we wrote this report, depicted unprecedented details about the altered immune further corroborating the claim. cell landscape in COVID-19 patients, increasing our The lymphopenia is another prominent feature of mechanistic and systemic understanding of COVID-19 immune perturbation of severe COVID-192,3,16. Here we immunopathogenesis, such as cytokine storm and revealed that the COVID-19-associated lymphopenia + lymphopenia. included not only depletion of CD8 T cells but also IFN production is stunted in SARS-CoV-2-infected cells significant loss of innate-like T cells, including MAIT, γδ and COVID-19 patients7,27. Consistent with these earlier T, and NKT cells, similarly reported by another study13,32. reports, here we found that, although ISGs in COVID-19 However, the proportion of cycling T cells is increased in patients were induced in monocyte-macrophages, neither COVID-19; whether this reflects SARS-CoV-2-specificT type I nor type III IFNs were produced. However, the cell response or bystander activation needs further higher levels of ISGs in monocyte-macrophages and investigation. In addition, we noticed that the frequencies + + T cells from mild cases than those in severe COVID-19 of CD4 T cells among all CD3 T lymphocytes were still supported a heightened IFN responses linked to increased, like CD4-GATA3 (Th2), CD4-CCR6 (Th17), resolving the diseases. It is unclear what causes the and CD4-ICOS (Tfh) cells. Indeed, these data are con- imbalance of inflammation and IFN production in sistent with earlier reports showing that lymphopenia in + + COVID-19; we assume that the loss of pDCs in severe severe COVID-19 affects CD4 T cells less than CD8 + cases may partly contribute to the diminished IFN pro- T cells33. CD4 T cell subsets also showed a tendency of duction. Notably, in contrast to their blood counterparts, higher clonal expansion, suggesting their activation status. monocyte-macrophages in paired BALFs produced higher We suspected that the disturbed T cell compartments levels of cytokines and chemokines, especially from those may contribute to COVID-19 immunopathogenesis, e.g., severe COVID-19 patients. Thus massive tissue-resident the impaired antiviral responses by innate-like T cells and + production of cytokines/chemokines and lack of IFN CD8 T cells, meanwhile amplifying inflammation and + induction suggest a crucial role played by local but not by inducing aberrant antibody responses by CD4 T cells. circulating monocyte-macrophages in fueling the cytokine Moreover, our transcriptional analysis dispute against the storm during severe COVID-19 and indicate that inhi- evidence of cytokine production, exhaustion, or increased biting the local cytokine storm might be more effective cell death by peripheral T cells in COVID-19 patients than those systemic intervention strategies. Moreover, the noted by several earlier studies26,34. Instead, current TCR roles of anti-inflammatory cytokines in COVID-19 tracking analysis suggested enhanced recruitment of + + pathogenesis, including IL1R2, IL1RN, IL10, and TGFB1, peripheral CD4 and CD8 T cells into lung tissues in are unclear and should be investigated in the future. COVID-19 patients, where they were induced to made Previously, several studies have reported that MHC II cytokines locally and likely contributed to cytokine storm molecules were downregulated in blood monocytes4,14,18 and peripheral lymphopenia. in patients with severe COVID-19; however, there was In conclusion, we comprehensively delineated the per- still no consensus on the upregulated genes. We found turbed immune landscapes during SARS-CoV-2 infec- here that genes related to neutrophil activation, including tions from both peripheral blood and infected lungs. Xu et al. Cell Discovery (2020) 6:73 Page 11 of 14

These data reveal potential cellular and molecular from peripheral blood were isolated by Ficoll-Hypaque mechanisms implicated in COVID-19 immunopathogen- density gradient centrifugation protocol. For subsequent esis and identify a peculiar functional dichotomy, per- study, the isolated cells were counted in 0.4% trypan blue, ipheral immune paralysis and broncho-alveolar immune centrifuged, and re-suspended at the concentration of 2 × hyperactivation, speciﬁcally in severe COVID-19. 106/mL.

Materials and methods Cytokines measurement by cytometric bead array Ethics statement Twelve cytokines, including IL-1β, IL-6, IL-8, etc., were This study was conducted according to the ethical detected according to the instruction (Uni-medica, principles of the Declaration of Helsinki. Ethical approval Shenzhen, China, Cat. No. 503022). In brief, the super- was obtained from the Research Ethics Committee of natant was taken from BALF after 10-min centrifugation Shenzhen Third People’s Hospital (2020-207). All parti- at 1000× g. Afterwards, 25 µL Sonicate Beads, 25 µL BALF cipants provided written informed consent for sample supernatant or plasma, and 25 µL of Detection Antibodies collection and subsequent analyses. were mixed and placed on a shaker at 500× rpm for 2 h at room temperature. Then 25 µL of SA-PE was added to Patients each tube directly. The tubes were then placed on a Thirteen COVID-19 patients were enrolled in this study shaker at 500× rpm for 30 min. The data were obtained by at the Shenzhen Third People’s Hospital. Metadata and flow cytometry (Canto II, BD) and were analyzed use patients’ samples were collected similarly as previously LEGENDplex v8.0 (VigeneTech Inc.). described8. The severity of COVID-19 was categorized to be mild, moderate, severe, and critical according to the Flow cytometry “Diagnosis and Treatment Protocol of COVID-19 (the 7th For cell-surface labeling, 1 × 106 cells were blocked with Tentative Version)” by National Health Commission of Fc-block reagent (BD Biosciences). Then the following China (http://www.nhc.gov.cn/yzygj/s7653p/202003/ antibodies were added and incubated for 30 min, includ- 46c9294a7dfe4cef80dc7f5912eb1989.shtml). In this ing anti-CD3 (BioLegend, HIT3a), anti-CD14 (BioLegend, study, we grouped patients with mild and moderate 63D3), anti-HLA-DR (BioLegend, L243), and anti-CD45 COVID-19 as the mild group and included those with (BioLegend, 2D1). After incubation, the samples were severe and critical diseases as the severe group. Three washed and reconstituted in phosphate-buffered saline for healthy subjects were enrolled as the control group. flow cytometric analysis on a FACSCanto II flow cytometer. Real-time quantitative PCR (qRT-PCR) for SARS-CoV-2 RNA In clinical practice, nasal swab, throat swab, sputum, scRNA-seq library construction anal swab or BALF could be collected for the SARS-CoV- The scRNA-seq libraries were prepared with the 2 nucleic acid assays. Total nucleic acid was extracted Chromium Single Cell VDJ Reagent Kits (10× Genomics; from the samples using the QIAamp RNA Viral kit PN-1000006, PN-1000014, PN-1000020, PN-1000005) (Qiagen) and qRT-PCR was performed using a China following the manufacturer’s instruction. Briefly, Gel bead Food and Drug Administration-approved commercial kit in Emulsion (GEM) are generated by combining barcoded specific for SARS-CoV-2 detection (GeneoDX). Each Gel Beads, a Master Mix containing 20,000 cells, and qRT-PCR assay provided a threshold cycle (Ct) value. The Partitioning Oil onto Chromium Chip B. Reverse tran- specimens were considered positive if the Ct value was ≤ scription takes place inside each GEM, after which cDNAs 37, and otherwise it was negative. Specimens with a Ct are pooled together for amplification and library con- value > 37 were repeated. The specimen was considered struction. The resulting library products consist of Illu- positive if the repeat results were the same as the initial mina adapters and sample indices, allowing pooling and result or between 37 and 40. If the repeat Ct was unde- sequencing of multiple libraries on the next-generation tectable, the specimen was considered to be negative. short read sequencer.

Immune cell isolation scRNA-seq data processing, cell clustering, and dimension For harvesting BALF cells, freshly obtained BALF was reduction placed on ice and processed within 2 h in the Biological We aligned the sequenced reads against GRCh38 Safety Level 3 laboratory. After passing BALF through a human reference genome by Cell Ranger (version 3.1.0, 100-µm nylon cell strainer to ﬁlter out cell aggregates and 10× genomics). To remove potential ambient RNAs, we debris, the remaining ﬂuid was centrifuged and the cell used the remove-background function in CellBender35, pellets were re-suspended in the cooled RPMI 1640 which removes ambient RNA contamination and random complete medium. For PBMC isolation, immune cells barcode swapping from the raw UMI-based scRNA-seq Xu et al. Cell Discovery (2020) 6:73 Page 12 of 14

data. Quality of cells were further assessed by the fol- upregulated if average natural logarithm fold change lowing criteria: (1) the number of sequenced genes is 200 (logFC) > 0.25 and adjusted P < 0.01. The genes with to 6000; (2) the total number of UMI per cells is > 1000; logFC < −0.25 and adjusted P < 0.01 were considered (3) the percentage of mitochondrial RNA is < 15% per cell. significantly down regulated. The selected genes shown Data integration, cell clustering, and dimension reduc- on the heatmap have logFC < −0.41 or > 0.41 and tion were performed by Seurat (version 3)36. First, we adjusted P < 0.01. identified 2000 highly variable genes (HVGs), which were used for the following analysis using FindVariableFeatures GO annotation function. Next, we integrated different samples by Inte- ClusterProfiler37 in R was used to perform GO term grateData function, which eliminates technical or batch enrichment analysis for the significantly upregulated and effect by canonical correlation analysis (CCA). Using downregulated genes. Only GO term of Biological Process those HVGs, we calculate a principal component analysis was displayed. (PCA) matrix with the top 50 components by RunPCA function. The cells were then clustered by FindClusters Calculation of composite score of MHC class II molecules function after building nearest neighbor graph using and calprotectin proteins FindNeighbors function. The cluster-specific marker Composite signature scores of MHC class II molecules + genes were identified by FindMarkers function using and calprotectin proteins of each peripheral CD14 MAST algorithm. The clustered cells were then projected monocytes were calculated using “AddModuleScore” into a two-dimension space for visualization by a non- function implemented in the Seurat package. MHC class linear dimensional reduction method RunUMAP in II score was calculated using the following genes, i.e., Seurat package. HLA-DMA, HLA-DMB, HLA-DPA1, HLA-DPB1, HLA- DQA1, HLA-DQB1, HLA-DRA, HLA-DRB1, and HLA- Integrated analysis of peripheral myeloid, NK, and T cells DRB5. Calprotectin protein score was calculated using For cells in PBMCs, we integrated the myeloid com- genes, including S100A1, S100A2, S100A3, S100A4, partment including mDC and monocytes or NK and S100A5, S100A6, S100A7, S100A7A, S100A7L2, T cells using the similar aforementioned procedure. We S100A7P1, S100A7P2, S100A8, S100A9, S100A10, re-clustered the myeloid or NK and T cells using the top S100A11, S100A12, S100A13, S100A14, S100A15A, 20 dimensions of PCA with default parameters. To obtain S100A16, S100B, S100G, S100P, and S100Z. The corre- high-resolution cell clusters for each subset, we set the lations between MDSC-like scores (“Calprotectin protein parameter resolution to 1.2. The cell clusters were score” minus “MHC class II score”) and plasma IL6, CRP annotated by canonical markers. levels, and the blood neutrophil/CD3/CD4/CD8 cell counts were calculated using GraphPad Prism 8.4.2. Lines Integrated analysis of myeloid, NK, and T cells from PBMCs were fitted using the simple linear regression method. and BALF Myeloid or NK and T cells from PBMCs and BALF were Pseudo-time trajectory analysis integrated separately. For myeloid cells, we extracted Trajectories analysis was performed using slingshot38 macrophage and mDC cells in BALF and monocyte and for monocyte-macrophages and T cells separately. For mDC in PBMCs from the corresponding raw count T cells in PBMC, naive CD4 and CD8 T cells were set as + + matrix. The extracted cells were integrated using CCA in the start point for CD4 T cells and CD8 T differ- Seurat (version 3) as mentioned above. For clustering, the entiation trajectory, respectively. For integrated analysis of resolution parameter was set to 0.6. Similarly, we monocyte-macrophages in PBMCs and BALF, we extracted NK and T cells in BALF and in PBMCs from the deduced the cell trajectory for each individual using the corresponding raw count matrix. The extracted cells were peripheral monocytes as the start point. integrated using CCA in Seurat (version 3) as mentioned above. For clustering, the resolution parameter was set Sc-TCR analysis to 1.5. The amino acid and nucleotide sequence of TCR chains were assembled and annotated by cellranger vdj function Analysis of DEGs in CellRanger (version 3.1.0). Only cells with paired TCRα FindMarkers function in Seurat (version 3) with MAST and TCRβ chains were included in clonotype analysis. algorithm was used to analyze DEGs. For each pairwise Cells sharing the same TCRα- and TCRβ-CDR3 amino comparison, we run FindMarkers function with para- acid sequences were assigned to the same TCR clonotype. meters of test.use=‘MAST’. The overlap of differentiated We assessed the TCR expansion, TCR transition among expressed genes among different comparisons was shown cell types, and TCR migration between PBMC and BALF by Venn diagram. Genes were defined as significantly using R package STARTRAC (version 0.1.0)39. TCR Xu et al. Cell Discovery (2020) 6:73 Page 13 of 14

migration between PBMC and BALF were shown using References the circos software40. 1. Guan, W. J. et al. Clinical characteristics of coronavirus disease 2019 in China. N. Engl.J.Med.382, 1708–1720 (2020). 2. Zhou, F. et al. Clinical course and risk factors for mortality of adult inpatients Statistics with COVID-19 in Wuhan, China: a retrospective cohort study. Lancet 395, The Student’s t-test (t.test in R, two-sided, unadjusted 1054–1062 (2020). 3. Chen,R.etal.Longitudinal hematologic and immunologic variations asso- for multiple comparisons) was used for pairwise com- ciated with the progression of COVID-19 patients in China. J. Allergy Clin. parisons of the cell proportions between different groups. Immunol. 146,89–100 (2020). Statistical difference of TCR expansion index and migra- 4. Xu, X. et al. Effective treatment of severe COVID-19 patients with tocilizumab. Proc. Natl. Acad. Sci. USA 117,10970–10975 (2020). tion index, between mild and severe disease group, were 5. Tay, M. Z., Poh, C. M., Renia, L., MacAry, P. A. & Ng, L. F. P. The trinity of COVID- calculated using t.test in R. Statistical difference of cyto- 19: immunity, inflammation and intervention. Nat. Rev. Immunol. 20, 363–374 kine level between BALF and blood in Fig. 3g were cal- (2020). 6. Roschewski, M. et al. Inhibition of Bruton tyrosine kinase in patients with culated using two-sided wilcox.test in R. severe COVID-19. Sci. Immunol. 5, eabd0110 (2020). 7. Blanco-Melo, D. et al. Imbalanced host response to SARS-CoV-2 drives Acknowledgements development of COVID-19. Cell 181,1036.e9–1045.e9 (2020). We acknowledge the work and contribution of all the health providers from 8. Liao, M. et al. Single-cell landscape of bronchoalveolar immune cells in Shenzhen Third People’s Hospital. This study was supported by the Science patients with COVID-19. Nat. Med. 26,842–844 (2020). and Technology Innovation Committee of Shenzhen Municipality 9. Bost, P. et al. Host-viral infection mapsrevealsignaturesofsevereCOVID-19 (2020A1111350032 to Z.Z.), the China Postdoctoral Science Foundation patients. Cell 181,1475.e12–1488.e12 (2020). (2020T130122ZX to F.Q.), and the National Natural Science Foundation of 10. Wu, F. et al. Evaluating the association of clinical characteristics with neu- China (81700540 to Y.L.). tralizing antibody levels in patients who have recovered from mild COVID-19 in Shanghai, China. JAMA Intern. Med. 180,1356–1362 (2020). Author details 11. Ni, L. et al. Detection of SARS-CoV-2-specific humoral and cellular immunity in 1Institute for Hepatology, National Clinical Research Center for Infectious COVID-19 convalescent individuals. Immunity 52,971.e3–977.e3 (2020). Disease, Shenzhen Third People’s Hospital, The Second Affiliated Hospital, 12. Grifoni, A. et al. Targets of T cell responses to SARS-CoV-2 coronavirus in School of Medicine, Southern University of Science and Technology, Shenzhen, humans with COVID-19 disease and unexposed individuals. Cell 181, 1489. Guangdong 518112, China. 2CAS Key Laboratory of Quantitative Engineering e15–1501.e15 (2020). Biology, Shenzhen Institute of Synthetic Biology, Shenzhen Institutes of 13. Wen, W. et al. Immune cell profiling of COVID-19 patients in the recovery Advanced Technology, Chinese Academy of Sciences, Shenzhen, Guangdong stage by single-cell sequencing. Cell Discov. 6,31(2020). 518055, China. 3Hepatology Unit, Department of Infectious Diseases, Nanfang 14. Giamarellos-Bourboulis, E. J. et al. Complex immune dysregulation in COVID- Hospital, Southern Medical University, Guangzhou, Guangdong 510515, China. 19 patients with severe respiratory failure. Cell Host Microbe. 27,992.e3–1000. 4Shanghai Public Health Clinical Center, Fudan University, Shanghai 201508, e3 (2020). China 15. Ildefonso Sanchez-Cerrillo, et al. Differential redistribution of activated monocyte and dendritic cell subsets to the lung associates with severity of Author contributions COVID-19. Preprint at https://doi.org/10.1101/2020.05.13.20100925 (2020). L.L., S.Z., and Z.Z. designed this study and wrote the manuscript. G.X. 16. Mathew, D. et al. Deep immune profiling of COVID-19 patients reveals distinct performed this study and wrote the manuscript. F.Q. performed the immunotypes with therapeutic implications. Science 369, eabc8511 (2020). bioinformatics analysis. H.L. and Y.L. revised the manuscript. Q.Y., X.W., X. Liu, 17. Chua, R. L. et al. COVID-19 severity correlates with airway epithelium-immune and J.Z. performed the experiments. H.W. and X. Liao collected the specimens cell interactions identified by single-cell analysis. Nat. Biotechnol. 38, 970–979 and provided the patients’ clinical information. (2020). 18. Wilk, A. J. et al. A single-cell atlas of the peripheral immune response in – Data availability patients with severe COVID-19. Nat. Med. 26,1070 1076 (2020). The raw data reported in this paper have been deposited in the Genome 19. Philip, A. M. et al. Targeted immunosuppression distinguishes COVID-19 from fl Sequence Archive in National Genomics Data Center, Beijing Institute of in uenza in moderate and severe disease. Preprint at https://doi.org/10.1101/ Genomics, Chinese Academy of Sciences under accession number(s) 2020.05.28.20115667 (2020). HRA000297 that are publicly accessible at http://bigd.big.ac.cn/gsa-human. 20. Weiss, P. & Murdoch, D. R. Clinical course and mortality risk of severe COVID- 19. Lancet 395,1014–1015 (2020). 21. Gabrilovich, D. I. & Nagaraj, S. Myeloid-derived suppressor cells as regulators of Code availability theimmunesystem.Nat. Rev. Immunol. 9,162–174 (2009). The source code and software pipeline to reproduce our analyses can be 22. Veglia, F., Perego, M. & Gabrilovich, D. Myeloid-derived suppressor cells assessed upon request. coming of age. Nat. Immunol. 19,108–119 (2018). 23. Mengos,A.E.,Gastineau,D.A.&Gustafson,M.P.TheCD14(+)HLA-DR(lo/neg) fl Con ict of interest monocyte: an immunosuppressive phenotype that restrains responses to fl The authors declare that they have no con ict of interest. cancer immunotherapy. Front. Immunol. 10, 1147 (2019). 24. Bronte, V. et al. Recommendations for myeloid-derived suppressor cell nomenclature and characterization standards. Nat. Commun. 7, 12150 (2016). Publisher’s note 25. Merad,M.&Martin,J.C.Pathologicalinflammation in patients with COVID-19: Springer Nature remains neutral with regard to jurisdictional claims in a key role for monocytes and macrophages. Nat. Rev. Immunol. 20, 355–362 published maps and institutional affiliations. (2020). 26. Yonggang, Z. et al. Pathogenic T cells and inflammatory monocytes incite Supplementary Information accompanies the paper at (https://doi.org/ inflammatory storm in severe COVID-19 patients. Natl Sci. Rev. nwaa041 (2020). 10.1038/s41421-020-00225-2). 27. Hadjadj, J. et al. Impaired type I interferon activity and inflammatory responses in severe COVID-19 patients. Science 369,718–724 (2020). Received: 26 August 2020 Accepted: 27 September 2020 28. Silvin, A. et al. Elevated calprotectin and abnormal myeloid cell subsets dis- criminate severe from mild COVID-19. Cell 182, 1401.e18–1418.e18 (2020). 29. Schulte-Schrepping, J. et al. Severe COVID-19 is marked by a dysregulated myeloid cell compartment. Cell 182, 1419.e23–1440.e23 (2020). Xu et al. Cell Discovery (2020) 6:73 Page 14 of 14

30. Arunachalam,P.S.etal.Systemsbiological assessment of immunity to mild vs from scRNA-seq datasets. Preprint at https://doi.org/10.1101/791699 severe COVID-19 infectioninhumans.Science 369,1210–1220 (2020). (2020). 31. PierreBost, F. et al. Deciphering the state of immune silence in fatal COVID-19 36. Stuart, T. et al. Comprehensive integration of single-cell data. Cell 177, 1888. patients. Preprint at https://doi.org/10.1101/2020.08.10.20170894 (2020). e21–1902.e21 (2019). 32. Jouan,Y.etal.Phenotypicalandfunctional alteration of unconventional T cells 37.Yu,G.,Wang,L.G.,Han,Y.&He,Q.Y.clusterProﬁler: an R package for in severe COVID-19 patients. J. Exp. Med. 217, e20200872 (2020). comparing biological themes among gene clusters. OMICS 16,284–287 33. Chen, Z. & John Wherry, E. T cell responses in patients with COVID-19. Nat. Rev. (2012). Immunol. 20, 529–536 (2020). 38. Street, K. et al. Slingshot: cell lineage and pseudotime inference for single-cell 34. Diao, B. et al. Reduction and functional exhaustion of T cells in patients transcriptomics. BMC Genomics 19, 477 (2018). with coronavirus disease 2019 (COVID-19). Front. Immunol. 11,827 39. Zhang, L. et al. Lineage tracking reveals dynamic relationships of T cells in (2020). colorectal cancer. Nature 564,268–272 (2018). 35. Fleming, S. J., Marconi, J. C. & Babadi, M. CellBender remove-background: a 40. Krzywinski, M. et al. Circos: an information aesthetic for comparative genomics. deep generative model for unsupervised removal of background noise Genome Res. 19,1639–1645 (2009).